LifeGuard AI: An Integrated Multi-Agent Framework for Longitudinal Health Record Management and Explainable Risk Assessment in Dependent Care

Effective healthcare for non-communicative dependents—infants, elderly individuals, andcompanion animals—remains a persistent challenge in clinical and home-care settings.Caregivers routinely receive fragmented physiological signals from heterogeneous wearableand sensor sources, yet lack transparent, trustworthy tools to interpret aggregate healthrisk or determine appropriate escalatory actions. Existing monitoring platforms either stopat raw data visualization or rely on opaque, single-model AI predictions that provide noexplanation and no safe pathway to action. This paper presents LifeGuard AI, an integrated framework combining two complementary subsystems: Mediforever, a longitudinal digital health record platform providing a lifelong, unified medical data layer; and AHRA (Agentic Health Risk Assessment), a multiagent orchestration system that analyzes physiological inputs, produces explainable risk scores, and safely coordinates real-world escalation workflows with explicit user consent. AHRA employs seven specialized agents—spanning context normalization, vital sign analysis, sleep evaluation, activity assessment, risk aggregation, threat classification, and action recommendation—orchestrated by a central Decision Orchestrator Agent. The system is intentionally data-source-agnostic, designed to ingest inputs from wearable sensors, IoTdevices, electronic health records, or manual entry. We present the system architecture, agent design rationale, a safety and ethics framework, and a pilot evaluation across five synthetic clinical vignette scenarios covering three dependent populations (human infants, canines, and exotic reptiles). Results demonstrate accurate risk stratification into four threat levels (Normal, Monitor, Warning, Critical), with 100% precision on critical condition detection across all test scenarios and mean system response latency of 3.2 seconds per assessment. A small expert review panel (n = 5) rated the explainability of agent outputs at 4.4/5.0. We discuss implications for AI in healthcare, limitations, and directions toward real-world deployment and clinical validation.